7 min read

Nearly every organization today has FinOps—a discipline for managing cloud costs. Yet companies continue to waste 25 to 35 percent of their cloud spending on unused resources. The figure has remained steady for years, even as tools improve. The issue isn’t a lack of visibility. It’s that visibility doesn’t equal the authority to shut something down.

Related:VMware VCF 9.1 and the sovereignty question  /  800-volt DC for AI racks

What FinOps is actually meant to achieve

What is FinOps? FinOps—short for Financial Operations—is the practice of making cloud costs transparent and managing them across engineering, finance, and business teams. The goal is to ensure every expense is tied to an owner and cost decisions are made where technical control resides.

In theory, FinOps closes an old gap. Before, cloud bills arrived at month-end, no one could assign costs to specific teams, and optimization was a guessing game. FinOps brings clarity: Which team incurs which costs? Which resources sit idle? Where do reservations make sense? The FinOps tool market has grown accordingly, and according to the State of FinOps Report, 98 percent of practitioners now also manage AI costs, while 90 percent track SaaS spending.

The tools exist—and they’re good. That’s what makes the persistent waste rate so striking. If visibility is improving but waste remains, visibility isn’t the bottleneck. The report pinpoints the issue, but reading it isn’t the same as acting on it.

Why the Number Isn’t Dropping

The reason is organizational, not technical. FinOps teams almost always have visibility, but rarely the authority. They can pinpoint exactly when a database has been idle for three months or an instance is three times oversized. What they can’t do: simply shut it down. Because the resource belongs to a development team that, if push comes to shove, insists it still needs it—and would be on the hook if something breaks.

This creates a stalemate. FinOps recommends, the team hesitates, no one decides. The unused resource keeps running because shutting it down carries a small risk, while letting it run only costs money that gets lost in the collective bill. Money that doesn’t hurt anyone personally will, in the end, always lose out to a risk someone would have to answer for. It’s human nature—and collectively expensive.

The AI wave sharpens this dynamic. GPU instances and inference endpoints cost many times more than a standard VM, and they’re often spun up for experiments that no one cleans up afterward. If no one shuts down an unused standard database, a forgotten GPU cluster is even less likely to be decommissioned. The 98 percent now managing AI costs are trying to control a line item that’s growing faster than the old stalemate can shrink it.

What Actually Lowers the Rate

The lever isn’t another tool—it’s authority. FinOps needs to shift from reporting to deciding, at least within clear rules. A proven approach is the mandate with a deadline: A resource flagged as unused is automatically shut down after a defined grace period unless the responsible team actively objects. The burden of proof flips. FinOps no longer has to justify shutting it down—the team has to justify keeping it running.

The second lever is assigning costs to the right budget. As long as cloud expenses land in a central IT pool, no individual team feels the impact. When costs are directly attributed to the team responsible, behavior changes automatically. What’s in your own budget gets shut down long before FinOps has to flag it.

A FinOps report without the authority to act is an expensive weather map. It predicts exactly where it’s raining, and no one gets an umbrella.

For AI resources, a third factor comes into play: an expiration date from the start. When someone spins up a GPU cluster for an experiment, they set a deadline for automatic shutdown. This prevents the most expensive form of waste—the forgotten high-performance resource. None of these three levers are technically difficult. All three are organizationally uncomfortable because they shift responsibility to where there was once only convenience.

That’s precisely why the number barely budges. Better tools solve a problem that was never about tools. If you want to cut those 25 to 35 percent, don’t buy another FinOps platform. Give the one you have a mandate. It’s the uncomfortable—but only—answer that’s actually moved the needle in recent years.

Frequently Asked Questions

Why isn’t cloud waste decreasing despite FinOps?

Because the problem is organizational, not technical. FinOps teams have the visibility, but rarely the authority to shut down a resource. As long as no one makes a decision, the unused resource keeps running.

How is AI impacting the cost landscape?

GPU and inference resources cost many times more than standard instances and are often spun up for experiments that no one ever cleans up. This exacerbates the exact category that was already the hardest to control.

How can FinOps gain the mandate to shut down resources?

Through a rule with a deadline: A resource flagged as unused is automatically shut down after a grace period unless the team objects. The burden of proof shifts from justifying the shutdown to justifying keeping it running.

Does reallocating costs help?

Significantly. As long as cloud costs sit in a central catch-all bucket, no team feels the impact of their consumption. When costs are directly attributed to the team causing them, shutdown behavior changes on its own.

Does this require new tools?

No. Existing FinOps tools already provide the necessary visibility. What’s missing is the authority and proper cost allocation. Adding another tool won’t solve a problem that was never a tooling issue to begin with.

Image source: AI-generated (June 2026), C2PA certificate embedded in the image